Video Quality Assessment and Reliable Object Descriptor for Video Surveillance Applications

Abstract

THIS HAS BEEN CANCELLED!

Intelligent solutions for a variety for applications in video surveillance require reliable assessment of video quality and detection of absolute attributes of objects of interest. Surveillance cameras tend to achieve a balance between the resolution of images captured and the rate at which video is collected. This trade-off imposes additional restrictions on the ability to identify objective image and object descriptors for effective vision tasks. In addition, most vision algorithms rely on the ability to detect required object features for reliable performance.

In this talk, we present an image quality metric that can determine the usability of a given image for a vision algorithm, independent of a reference. This no-reference image quality measure attempts to capture effects of image blurriness, blockiness and aliasing on performance of vision algorithms. Such a quality measure can be used to evaluate the input image to a vision algorithm so as to optimize an algorithm’s probability of success. A mechanism to combine the image quality measures of individual video frames into a global video quality measure is also proposed. The effectiveness of this image/video quality measure is evaluated by its application to face detection in video sequences.

For an absolute measure of object attribute, a reliable estimate of the object height from any uncalibrated camera can prove to be a useful descriptor. Several approaches have been developed to obtain height information of an object from image sequences, but are limited in their utility due to errors arising from varying image and object properties. We present a simplified error model dependent on the location of the object in the image and the estimated camera height. Further, the developed model is used to obtain reliable height estimates of objects in different image sequences and the result compared to estimates obtained without the use of the model. Significant reduction in error associated with the measurement is observed while making the estimate independent of camera and object variations.

Speaker Biography

Shishir Shah is Assistant Professor of Computer Science at the University of Houston. He received his M.S. and Ph.D. degrees in Electrical and Computer Engineering from The University of Texas at Austin. He received his B.S. degree in Mechanical Engineering from The University of Texas at Austin. His current research includes computer vision and pattern recognition with applications in biomedical image analysis and distributed multi-modality sensing. He has co-edited one book, and authored papers on object recognition, sensor fusion, statistical pattern analysis, and bioinformatics.